(Not) Understanding Latin Poetic Style with Deep Learning
This addresses the problem of interpreting neural networks for literary analysis, though it's incremental as the main goal of understanding style remains unachieved.
The researchers attempted to understand authorial style in classical Latin poetry using neural networks (LSTMs and CNNs) trained on encoded sonic and metrical features, achieving strong authorship classification but finding the models' reasoning inscrutable. They found CNNs train faster with equivalent accuracy and potentially better interpretability than LSTMs, and that simple trainable embeddings outperform domain-specific schemes.
This article summarizes some mostly unsuccessful attempts to understand authorial style by examining the attention of various neural networks (LSTMs and CNNs) trained on a corpus of classical Latin verse that has been encoded to include sonic and metrical features. Carefully configured neural networks are shown to be extremely strong authorship classifiers, so it is hoped that they might therefore teach `traditional' readers something about how the authors differ in style. Sadly their reasoning is, so far, inscrutable. While the overall goal has not yet been reached, this work reports some useful findings in terms of effective ways to encode and embed verse, the relative strengths and weaknesses of the neural network families, and useful (and not so useful) techniques for designing and inspecting NN models in this domain. This article suggests that, for poetry, CNNs are better choices than LSTMs -- they train more quickly, have equivalent accuracy, and (potentially) offer better interpretability. Based on a great deal of experimentation, it also suggests that simple, trainable embeddings are more effective than domain-specific schemes, and stresses the importance of techniques to reduce overfitting, like dropout and batch normalization.